NVIDIA Releases Nemotron 3.5 Super VL and Bolt Models for Edge and Enterprise Agents

NVIDIA released Nemotron 3.5 Super VL, an open 120-billion-parameter model designed for specialized always-on agents, alongside two new edge models: Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB. These additions expand the Nemotron family, which already includes Lightning and Ultra, to cover deployments from large enterprise servers to local hardware and edge devices.

Nemotron 3.5 Super VL

Super VL delivers frontier-level accuracy on multimodal agentic tasks, including instruction following, tool use, and visual understanding, comparable to NVIDIA’s largest reasoning models at roughly one-fifth the parameter count. It is the fastest open agentic model in its class, achieving up to 5x faster throughput than competitors on the Artificial Analysis Intelligence Index. The model was trained on approximately 26 trillion tokens and post-trained with roughly 3.8 million samples.

Organizations can post-train Super VL on their own data using NVIDIA NeMo and open datasets. Adoption is already underway, with companies like Scale AI, Atlassian, CrowdStrike, and Perplexity customizing the model for specific workflows such as financial analysis, agentic search, and cybersecurity. Super VL is available now on Hugging Face, Open Router, ModelScope, and as NVIDIA NIM.

Nemotron 3.5 Bolt Edge Models

The Bolt models target local and edge environments. Bolt 74B-A9B 24GB uses native 2-bit quantization to fit into 24GB of memory, allowing it to run on NVIDIA RTX/DGX Spark systems. It ranks #1 on PinchBench and #2 on BrowseComp among open models in its class. Bolt E4B 4GB, optimized for intelligence per GB, is designed for physical and conversational agents on GeForce RTX, Jetson, and RTX/DGX Spark hardware.

Bolt 74B-A9B is available now on Hugging Face with support for llama.cpp, Ollama, LM Studio, Unsloth, and vLLM. Bolt E4B is scheduled for release on Oct. 23.